Employee Performance Evaluation Model Based on Digital Footprints Using Machine Learning

Abstract

Employee performance evaluation is a crucial aspect of human resource management, yet traditional methods are often subjective and less effective. This study develops an employee performance evaluation model based on digital footprint using machine learning algorithms to provide a more objective and fair assessment. Data were collected from 2,435 employees, including 16 input variables related to digital work activities, wearable and health data, work perception and satisfaction, competencies, and organizational information, with two output variables: Performance_Score and Performance_Label. The research stages included feature selection using ANOVA F-test, data preprocessing (missing value imputation and normalization with RobustScaler), class balancing using SMOTE, splitting the dataset into training (2,688 samples) and testing (672 samples) sets, and training models with Logistic Regression, Decision Tree, Random Forest, Gradient Boosting, KNN, SVM, and Naïve Bayes. Model evaluation was performed using accuracy, precision, recall, F1-score, and confusion matrix metrics. The results show that Random Forest achieved the best performance with an accuracy of 0.872 and F1-score of 0.871, followed by Gradient Boosting (accuracy 0.830, F1-score 0.827) and Decision Tree (accuracy 0.826, F1-score 0.822). KNN and SVM exhibited moderate performance, while Logistic Regression and Naïve Bayes had the lowest results. The most influential features were Organization Support Score, Tech Skills Score, Physical Activity Steps, Daily Working Hours, and Overtime Hours Monthly. The machine learning-based model, particularly Random Forest, is effective for employee performance evaluation based on digital footprints, enabling more objective, fair, and accurate assessments.